Fix mapreduce over dimensions with heterogeneous eltype (#28089)

The existing check does not make sense in now. It originates in a generalization
of a method which worked only for + (54e034a). Add a test for this based on missing,
and improve existing tests.

base/reducedim.jl

@@ -110,7 +110,7 @@ function _reducedim_init(f, op, fv, fop, A, region)
     if T !== Any && applicable(zero, T)
         x = f(zero(T))
         z = op(fv(x), fv(x))
-        Tr = typeof(z) == typeof(x) && !isbitstype(T) ? T : typeof(z)
+        Tr = z isa T ? T : typeof(z)
     else
         z = fv(fop(f, A))
         Tr = typeof(z)

test/reducedim.jl

@@ -106,9 +106,21 @@ end
 @test typeof(@inferred(Base.prod(abs, [1.0+1.0im], dims=1))) == Vector{Float64}
 @test typeof(@inferred(Base.prod(abs2, [1.0+1.0im], dims=1))) == Vector{Float64}
 
-# Heterogeneously typed arrays
-@test sum(Union{Float32, Float64}[1.0], dims=1) == [1.0]
-@test prod(Union{Float32, Float64}[1.0], dims=1) == [1.0]
+@testset "heterogeneously typed arrays" begin
+    for x in (sum(Union{Float32, Float64}[1.0], dims=1),
+              prod(Union{Float32, Float64}[1.0], dims=1))
+        @test x == [1.0]
+        @test x isa Vector{Float64}
+    end
+
+    x = sum(Real[1.0], dims=1)
+    @test x == [1.0]
+    @test x isa Vector{Real}
+
+    x = mapreduce(cos, +, Union{Int,Missing}[1, 2], dims=1)
+    @test x == mapreduce(cos, +, [1, 2], dims=1)
+    @test x isa Vector{Float64}
+end
 
 @test reduce((a,b) -> a|b, [true false; false false], dims=1, init=false) == [true false]
 let R = reduce((a,b) -> a+b, [1 2; 3 4], dims=2, init=0.0)